Recent Change? /PAE with 32GB of hot swap RAM

Discussion in 'Windows Server' started by Blunderdog, Jan 14, 2008.

  1. Blunderdog

    Blunderdog Guest

    For a little over a year we've run a cluster with Server 2003 32bit machines
    with quad processors and 32 GB of hot swapable ram. According to
    http://support.microsoft.com/kb/283037 the /PAE switch is not required in the
    boot.ini for applications that can use the ram to get access to it. In this
    case it is SQL 2000. We've run fine without that switch since September 2006.
    Today we noticed these machine were only reporting 3.5 GB of RAM in system
    manager. However when the machines boot the system clearly reports all 32 GB.

    To fix the problem I added the /PAE switch and rebooted the passive node and
    now Windows reports 32GB. Has something changed, perhaps a recent update,
    that has changed or required this behavior. We last performed updates on
    12/28.
     
    Blunderdog, Jan 14, 2008
    #1
    1. Advertisements

  2. Blunderdog

    Ryan Hanisco Guest

    Hi Blunderdog,

    I have not seen any updates that would change this behavior. As it stands
    the /PAE switch would not hurt you even if you didn't have hot-swappable
    memory and I would include it to be sure that there was no confusion in the
    memory space addressing. That being said, there are some possible
    differences between hardware vendors and it is possible that there was a
    detection issue with you specific server/ MB/BIOS/Memory configuration that
    would not have it detect it as hot-swappable -- though I would have expected
    this to have affected you all along, unless you just did BIOS updates.

    If you are sure that you always had access to the memory (remember that the
    boot declaration of memory is different from what the OS thinks it has) then
    I would consider the situation highly suspect. I would go back through he
    logs looking for BIOS or firmware updates and monitor the memory for issues.
    If this is part of a cluster, I would consider taking the node offline and
    running a memory diagnostic to be sure that there are no memory problems
    lurking.

    Other than that, I would leave the /PAE switch in and call it good.
    --
    Ryan Hanisco
    MCSE, MCTS: SQL 2005, Project+
    http://www.techsterity.com
    Chicago, IL

    Remember: Marking helpful answers helps everyone find the info they need
    quickly.
     
    Ryan Hanisco, Jan 15, 2008
    #2
    1. Advertisements

  3. Blunderdog

    Blunderdog Guest

    I will check to see if bios upgrades were done recently. That is not out of
    the question since I know some HBA updates and other stuff was done recently
    as well. If so then that will be the explanation.

    The full 32 GB of RAM had been showing in both the OS, BIOS and SQL since we
    got it all configured properly. We just noticed yesterday that this was no
    longer the case.

    I like your plan, for now, the /PAE switch has restored things to normal. I
    got all nodes rebooted last night.
     
    Blunderdog, Jan 15, 2008
    #3
    1. Advertisements

Ask a Question

Want to reply to this thread or ask your own question?

You'll need to choose a username for the site, which only take a couple of moments (here). After that, you can post your question and our members will help you out.